A holistic view
for machine
vision
By Arnaud Destruels, MV Product
Marketing Manager, Sony Image
Sensing Solution Europe
E
veryone wants
to achieve high
performance in
machine vision,
but it is a multifaceted
problem that demands a
deep understanding of how
image processing functions
work and knowledge about
the rate-limiting factors that
generally afflict machine
vision.
In an ideal world, the
illumination we have in
a machine vision system
would allow consistent
exposure of every part of
the image. But even in
the carefully controlled
conditions of a productionline this is difficult to
achieve. The surfaces we
are trying to capture may
be highly reflective in some
places and offer low contrast
in others. The result is glare
in one portion with indefinite
shapes in another.
The cameras used within
the machine vision industry
often contain features
which are suitable for nonmanufacturing activites.
In the case of intelligent
transportation systems
(ITS), there is often less
control over lighting. The
movement of the sun can
obscure number plates, for
example vehicles passing
each other on a highway
could lead to the shading
the number plates and
other key objects that the
system needs to recognise.
There is very little we
can do about this short of
building a highly distracting
lighting system into the
environment.
The key approach is to
take a step back and see
how we can apply imaging
technologies such as
CMOS to the problem. One
possibility is to leverage
the high throughput of
CMOS imagers to improve
overall image quality. CMOS
imagers support massive
internal parallelism, with
an A/D converter at the
end of each row. This
makes it possible to read
out pixel data at much
higher speeds than is
possible with traditional CCD
architectures.
By increasing the sensitivity
of the core CMOS pixel
and leveraging the parallel
architecture, it is now
possible to build imagers
that can capture 150 frames
per second. With those
levels of performanc e,
it is possible to combine
multiple frames into a
single composite image
to remove defects caused
by factors such as heat
haze. By subtly adjusting
the exposure of each
frame in the composite,
we can provide images
with much higher dynamic
range than is possible with
single captures. With a
high dynamic range image,
parts of the image can be
adjusted for brightness and
contrast without sacrificing
overall accuracy. The result
is much greater consistency
under changing, difficult
lighting conditions.
But we also need to
understand how the core
architecture of the CMOS
imager can potentially work
against us. Consumer video
cameras generally use
a rolling-shutter design,
where each pixel is captured
at a slightly different time to
its neighbour. In industrial
imaging, we cannot afford
the effects of that time
distortion. So, the answer
is to apply a global shutter
approach to ensure that
each row within a frame is
captured at precisely the
same time.
The performance
advantages offered
by high capture rates
and global shutter
technology need
to go hand in hand
with other systemdesign choices. Spatial
accuracy is as vital as
temporal accuracy. So,
Sony has designed its GS
CMOS cameras to support
extremely fine mounting
54
PECM Issue 18
tolerances to ensure that
image capture remains
highly consistent.
Support for protocols such
as the IEEE1588 Precision
Time Protocol further
enhances accuracy in
production line systems by
ensuring that the software
can identify each part
uniquely as it passes the
camera and signal where
it is on the line if it needs
to be removed or further
processed by a robot
downstream. At the same
time, bottleneck removal
techniques such as collision
avoidance in the networking
interface ensure that the
system is not adversely
affected by data congestion.
All of these factors, and
others, need to come
together to build a highperformance imaging
system. It’s not simply
about increasing pixel
counts and frame rates,
although those remain
important components of
the overall solution.
The performance
advantages offered by high
capture rates and global
shutter technology need to
go hand in hand with other
system-design choices.
Spatial accuracy is as vital
as temporal accuracy. So,
Sony has designed its GS
CMOS cameras to support
extremely fine mounting
tolerances to ensure that
image capture remains
highly consistent.